Methodology
The Shortlist asks the buying questions people now put to AI and publishes the answer as a clean, crawlable instrument. The spine is deliberately simple: one real ChatGPT call per question, shown in full, plus a derived layer that costs nothing to grow.
81 shortlists · 448 tools · 136 head-to-heads · asked via ChatGPT
How we ask
Each shortlist is one ChatGPT answer to the exact question a buyer would type, asked with web search off so the answer is the model’s trained judgment rather than whatever it could scrape in the moment. The model returns a strict JSON object: a ranked list of real products, each with a one-liner, who it’s for, what stands out, what to watch, a hedged price and a tier, and we store it.
What’s AI judgment, and what’s derived
| Layer | Source |
|---|---|
| The ranking and its order | ChatGPT |
| One-liners, best-for, stands-out, watch-out | ChatGPT |
| Pricing beliefs and confidence | ChatGPT |
| Tier tags and decision guide | ChatGPT |
| The snub and contrarian pick | ChatGPT |
| Slugs, domains, SEO strings | Derived |
| Entity, compare and alternatives pages | Derived |
| Overview, how-to-choose, clustering + extra FAQs | Derived |
The aggregation layer (zero calls)
The product profiles, head-to-heads and alternatives aren’t asked. They’re a roll-up over the whole corpus. Grouping a tool’s appearances gives its average rank, its wins, who it beats and loses to, and the tools AI names alongside it, all for free.
The honesty rules
- We publish the whole answer, including the pricing and recency the model is unsure about.
- We name the snub. When the model leaves off a well-known tool, we say so by name.
- We mark every estimate. Prices are the model’s belief, not a quote.
- We never fake other models. Default pages name ChatGPT specifically; real multi-model consensus is reserved for the clearly-labelled Consensus pages.